[GE users] MPICH2 & SGE tight integration

Reuti reuti at staff.uni-marburg.de
Mon Jan 2 23:30:06 GMT 2006


    [ The following text is in the "WINDOWS-1252" character set. ]
    [ Your display is set for the "ISO-8859-10" character set.  ]
    [ Some special characters may be displayed incorrectly. ]

Hi Ray,

the daemonless smpd doesn't need a startup of the smpds by hand. The  
behavior is just like the rsh startup from the original mpich. You  
should also see in the "ps -e f" output that the processes are just  
spawned to the nodes, without involving the smpds.

Did you got errors if started without starting the smpd first?
Did you compile your MPI program with the smpd version for daemonless  
startup?

As outlined in the Howto, you will get a different mpich2 library for  
each startup method and also a different mpiexec for each one. And  
the actual mpiexec used to start your program must match the compiled- 
in library in your MPI program.

Cheers - Reuti


Am 03.01.2006 um 00:04 schrieb Raymond Chan:

> Hi all,
>
>
>
> I followed Reuti?s tight integration procedure for daemonless  
> smpd.  With a single user, I must start smpd on the nodes (smpd ? 
> s), and my app runs fine.  As soon as I try to run another mpich2  
> job for more than this one user, I get errors.  Doing a smpd ?s on  
> another user after the first, gives me errors like this:
>
>
>
> [00:14377]..ERROR:MPIDU_Sock_listen failed,
>
> sock error: generic socket failure, error stack:
>
> MPIDU_Sock_listen(399): unable to bind socket to port  
> (port=6029704,errno=98:Address already in use)
>
>
>
> It seems that the first user?s smpd supersedes every other users?,  
> and subsequent user startups of smpd?s assumes the same port  
> binding as the original user who started it.  I then tried to start  
> smpd on a different port for each user w/ the ?p option or putting  
> it in the  .smpd file, but the port always seems to default back to  
> that 6029704.  I?m trying to get a multi-user smpd/mpich2  
> environment working w/ Sun Grid Engine.  Has anyone experienced  
> such problems getting things to run w/ more than one user?
>
>
>
> Thanks,
>
> -Ray C
>
>  Univ of California, Davis
>
>


---------------------------------------------------------------------
To unsubscribe, e-mail: users-unsubscribe at gridengine.sunsource.net
For additional commands, e-mail: users-help at gridengine.sunsource.net




More information about the gridengine-users mailing list